In the 1960s, Silicon Valley grew out of Stanford University and the semiconductor firms that emerged from Bell Labs. Since then, its centre of gravity has shifted epoch over epoch, from a foundational focus on hardware, to the architectures of global networks, to software and internet platforms. It has now violently swung back to hardware and the massive, specialised clusters that underpin the development of artificial intelligence.
The objective of each such shift, however, has remained unchanged: to improve the capacity to collect, or appropriate, and make use of information, or rather, data.
Beneath their veneer of vitality, LLMs, and other generative architectures, remain sophisticated mechanisms for recompiling and repurposing data, dependent upon the information they can harvest for training and later access at inference.
As these models improve, it becomes increasingly clear that there is no ghost, just a shell.
This is not to deny their power, it is simply to say—and this may be obvious—that we are seeing the emergence of data as a factor of production in the strictest economic sense, alongside, and progressively encompassing, land, labour, and capital. More than this, it constitutes the vast attack surface vulnerable to misaligned artificial intelligence, and proves to be, as attempts to meaningfully synthesise it fail, irreducibly real.
What is peculiar is that, despite this, we still possess no civil mechanism to govern data’s acquisition or even a general market architecture for its exchange; only brokers for bulk licences and training agreements acting as its “bucket shops” of the 21st century. Its unchecked accumulation remains a land grab whose only historical equivalent is Enclosure. This is not only an ethical failure but a structural barrier to the technology's own development.
No moral appeal from the Vatican, monkey-wrenching of data centres, moratoria on their development, boycotting of Palantir, or violence against any tech executive, will solve this.
Neither will Bernie Sanders’ demand for the redistribution of corporate equity, nor even Sam Altman’s dubious offer to give it (to make the public shareholders of their own cognitive disenfranchisement: to profit, marginally, from the expropriation of one’s own intellect).
In other words, there seems no framework within America’s traditions of democratic capitalism or socialism equipped to answer the existential question of this century: under what conditions can intelligence become common infrastructure?
Achieving it requires a structural transformation as profound as the Reformation: a deep technical and social challenge to an elite whose political economy of information has lost its legitimacy and, it cannot deny, failed.
This would be a computational architecture of what should be called public intelligence.
The purpose of this memo is to introduce what we believe preliminary aspects of that architecture to be. We begin by outlining the structural enclosure of data (1–52): why the compounding liability of holding data paralysed civic institutions, and will soon paralyse models too.
We then advance five components of public intelligence:
a separation of powers for computation (53–61): Google built a private architecture over public information; we propose a public architecture for private information. This framework would separate who holds data from who computes over it, which the nascent field of programmable cryptography technically enables; we believe this to be the kernel of an end to surveillance capitalism, and of a social utility, and economic power, of data not available at present. P4PI governs the inputs, not the models nor what they produce.
markets in computational rights and data (62–84): to realise this new paradigm requires a market infrastructure, built upon what we term "computational warrants". The warrant changes the basic economic transaction from "you may have access to my data" to "you may use my information to do this particular thing that we have agreed"; this, in turn, requires an equitable framework for the validation and acquisition of data, that users can permission. P4PI governs the inputs and the terms, not the models and not what they produce.
a reinvention of the public university (86–117): The historical triumph of Silicon Valley, with Stanford University, was to fuse federal research, intellectual property, finance, and recruitment into an extraordinarily productive engine of technological development. The university is a complex information-sharing structure and the root of much of the specialised knowledge on which AI models train; there is a practical opportunity for its public form to be rebuilt as the anchor of this new architecture of computation and intelligence. As it stands, the public university is also a hinge for K–12 education, libraries, and research hospitals, and is where data sanctuaries can be established as a foundational right across civic life. That begins with children, whose neurological development must not be captured by proprietary product suites, whether Anthropic’s, Google’s, or McGraw Hill’s.
public patent pools and open silicon (118–130): “open-source” is a strategic decoy so long as the physical means of computation remain proprietary. Underpinned by the public university system, we must create patent pools that extend open-source principles down to the chips themselves, the foundation that the new data economy and new kinds of computation require. Mirroring the 1956 AT&T antitrust consent decree—which compelled the royalty-free licensing of Bell’s patents, the transistor among them, and birthed Silicon Valley—this legal mechanism will open the proprietary basis of chip development, optimising them for the separation of powers. There is no other way to prevent a monopoly on the knowledge industry.
cooperative infrastructure (131–157): Frontier development is encountering a hard floor dictated by the capital costs of silicon and the physical limits of energy; as we know, data-centre expansion is driving unprecedented grid deficits and price increases, costs that land unevenly, on rural America. Drawing on Roosevelt's Rural Electrification Administration and Adriano Olivetti's comunità concreta, computational capacity can instead be tethered to local, cooperative ownership, which, together with the public university, offers the possibility of a new consensus between rural and urban America. The same cooperatives, under a separation of powers, would hold sensor networks and public structures for encryption; otherwise Flock cameras are only a harbinger of the private surveillance networks that will flood America. As ever, American legitimacy depends upon widening access to the productive capacities of its age.
At the close, we ask what political orientations follow, in a world where configurations of hardware and software increasingly constitute political identities.
To be absolutely clear: P4PI is neither a plan for a national AI model, nor a government body to control intelligence; it is policy for democratically determining how human knowledge forms synthetic intelligence, rivalling what is at present an intelligence-industrial complex.
The revelation we are stumbling into is that intelligence is, and always has been, a public good—and that to care for it is the most human of activities. As intelligence, in turn, becomes literally infrastructural, it also becomes the most material of political questions.
Our politics is one of the affordability of, and equitable access to, housing, watts, tuition, and tokens—these will be settled together or not at all, a question of democratic devolution and progressive technology policy.